We have no tests that this fails for, but on an EBCDIC machine, the
branches here are incorrect. They are trying to determine if a UTF-8
representation will be larger than a non-UTF-8 representation for code
points < 256. The proper test is if the code points are UTF-8
invariant.
tbl[i] = (short)i;
}
else {
- if (i < 128 && r[j] >= 128)
+ if (UVCHR_IS_INVARIANT(i) && ! UVCHR_IS_INVARIANT(r[j]))
grows = 1;
tbl[i] = r[j++];
}
--j;
}
if (tbl[t[i]] == -1) {
- if (t[i] < 128 && r[j] >= 128)
+ if ( UVCHR_IS_INVARIANT(t[i])
+ && ! UVCHR_IS_INVARIANT(r[j]))
grows = 1;
tbl[t[i]] = r[j];
}